CasFinder System Documentation: Strategy, Installation, Usage, and Performance Supporting Information for “CasFinder: Flexible algorithm for identifying specific Cas9 targets in genomes” John Aach, Prashant Mali, George M. Church* Department of Genetics, Harvard Medical School, Boston, MA 02115, USA * Corresponding author Contents CasFinder design ……………………………..………………………………………………………………………………………….. 3 CasFinder design overview …………………………………………………………………………………………………………. 3 Technical note: Why queries combining targeting sequence and PAM are difficult …………………… 4 Overview of the CasFinder system of programs ……………………………………………………………………………. 6 CasFinder system availability and citation …………………………………………………………………………………… 7 CasFinder system installation and customization ………………………………………………………………………… 8 1. Installing CasFinder program components……………………………………………………………………………… 8 2. Preparing genome files needed for CasFinder operation ……………………………………………………….. 8 3. Editing the CasFinder system configuration files ……………………………………………………………………. 9 4. Testing the installation ………………………………………………………………………………………………………….. 13 General features of CasFinder and CasValue operation ………………………………………………………………. 14 Program option attributes and coordination …………………………………………………………………………… 14 Input sequence processing options …………………………………………………………………………………………. 17 CasFinder operation …………………………………………………………………………………………………………………….. 18 CasFinder program options………………………………………………………………………………………………………... 18 CasFinder input file specifications ……………………………………………………………………………………………… 20 CasFinder output files ……………………………………………………………………………………………………………….. 22 CasValue operation ……………………………………………………………………………………………………………………… 26 CasValue program options ………………………………………………………………………………………………………… 26 CasValue input file specifications ………………………………………………………………………………………………. 28 CasValue output files ………………………………………………………………………………………………………………… 30 Sensitivity, specificity, and FDR of specific Cas9 site detection: Preliminary assessment …………….. 33 Using CasValue to re-screen CasFinder targets for higher specificity ……………………………………………. 36 CasFinder system performance considerations …………………………………………………………………………… 40 Generation of the all human and mouse exome catalogs of Cas9 sites ……………………………………….. 42 Definition of the human and mouse exomes …………………………………………………………………………... 43 Generation of the Cas9 site catalogs for each exome ……………………………………………………………….. 44 Generation of reduced set of Cas9 targets for gene knockout screens …………………………………….. 44 Statistics at the gene copy level ………………………………………………………………………………………………… 46 Comparison of the CasFinder system with other Cas9 target evaluation algorithms …………………… 46 References …………………………………………………………………………………………………………………………………… 49 1 Tables and Figures Table S1: Components of the CasFinder system of programs ……………………………………………………… 7 Table S2: CasFinder program options …………………………………………………………………………………………… 19 Table S3: CasValue program options …………………………………………………………………………………………… 28 Table S4: Cas9-targeted sites characterized for off-targets in human genome …………………………… 34 Table S5: Use of CasValue to screen sites for higher specificity ……………………………………………………. 38 Table S6: Numbers of sites accepted at default settings rejected at smaller seed lengths ………….. 40 Table S7: Numbers of gene copies with targets that do not appear in reduced target sets …………. 46 Table S8: Comparison of CasFinder system with other Cas9 target evaluation algorithms ………….. 47 Figure S3: Excerpt of CASFINDER_CONFIG.txt file at authors’ installation …………………………………… 10 Figure S4: Excerpt of distributed CASFINDER_CAS9_CONFIG.txt …………………………………………………. 13 Figure S5: Fraction of inaccurately evaluated SP Cas9 sites at different bowtie –k values …………… 24 (Note: Figures S1 and S2 are provided in separate Supporting Information documents.) 2 CasFinder design CasFinder design overview As described in the main article, in type II prokaryotic Clustered Regularly Interspaced Palindromic Repeats (CRISPR) systems, a complex of a Cas9 protein with a pair of RNAs recognizes a DNA target that consists of a short stretch of DNA complementary to the 5’ end of one of the RNAs followed by a Cas9- specific Protospacer Adjacent Motif (PAM). In applications, the two RNAs are frequently combined into a single guide RNA (sgRNA) whose 5’-most 20bp comprises the targeting region. The initial characterization of Cas9 specificity suggested that mismatches between the last 12-13nt of the RNA targeting region (the “seed” region) and DNA targets would abrogate Cas9 activity [4]. This led to the following simple and efficient algorithm for identifying putatively specific Cas9 targets in a genome: For any given candidate 20bp sequence followed by a PAM motif, all one need do is determine whether there is an exactly matching seed sequence followed by a PAM in the genome elsewhere than the candidate itself, and reject any candidate for which one is found as ‘non-specific’. In our initial work in ref. [3], we rapidly and easily implemented this algorithm for S. pyogenes (SP) Cas9, by extracting all 20nt candidate Cas9 target sites 5’-NNNNN NNBBB BBBBB BBBBB-NGG-3’ within the sequences of interest (where the B(13) region was taken as the seed sequence within the site and the final NGG is the SP Cas9 PAM), and using bowtie [5] to find exact matches in the genome to the four query sequences B(13)AGG, B(13)CGG, B(13)GGG, and B(13)TGG. The need for four queries arose because bowtie does not support ‘wildcard’ N bases in queries, but performance was still very good because bowtie is extremely fast. Note that while the length of the seed region is not settled, for practical reasons described in the article text we used 13bp seeds for human genome queries (and continue to do so here). However, the present understanding of specificity requires that seed mismatches, non-seed matches, and multiple and sometimes complex PAMs be taken into account (see main article for references), and this greatly complicates this strategy (for details, see the Technical note below). Thus, in our present method, we forewent the performance advantages of combined seed and PAM queries, in favor of a simple, practical, and highly flexible system with the following design: 1. The system implements two distinct functions: (i) finding candidate Cas9 sites within user- specified sequences, performed by the CasFinder program, and (ii) evaluating candidate target sequences in a genome, performed by CasValue. CasFinder calls CasValue to evaluate the candidates it finds, but CasValue may be called independently and can be used to evaluate the specificity of Cas9 targets found by other systems. 2. Because an increasing number of Cas9s are coming into use and uncertainty still exists regarding the structural features of their Cas9 sites, Cas9 target definitions are maintained entirely in user- editable configuration files separate from program code. Cas9s defined to the system have the following properties: (i) a target sequence length, (ii) a start and end point of a seed region within the target sequence, (iii) a list of possible primary PAM sequences that are specifiable using standard degenerate base codes, (iv) a similar list of possible secondary PAM sequences. 3. CasValue evaluates lists of input target sequences of the form N(L) = 5’-N(u)B(s)N(d)-3’ for the presence of potential off-targets in a genome, where L is the target length and s is the seed length defined for the Cas9 it is processing, and N(u) and N(d) are upstream and downstream non-seed sequences. CasValue starts by (i) using bowtie to find all matches to B(s) seed sequences in a specified genome with up to a specific number of mismatches, and then it (ii) retrieves from the genome sequence the full 5’-N’(u)B’(s)N’(d)-P(k)-3’ Cas9 footprints around each match, where k is the Cas9’s maximum PAM length. (iii) Matching of the P(k) against 3 primary or secondary PAMs is checked via regular expression pattern matches, and sites for which no match is found are discarded. (iv) For remaining footprints, a score is computed as a weighted sum of the numbers of mismatches between the two N’ and N regions, and of the B’ and B regions (where different weights may be applied to the N and B regions), and an additional cost is added if the P(k) matches a secondary but not a primary PAM. (v) CasValue is further provided an exclusion parameter –x that defines the upper limit of scores of matches it will consider as potential off-targets, and ignores scores > –x. (v) Finally, each input target sequence has an ‘accept’ number that indicates the number of genome matches with score ≤ −x that the input sequence may be allowed to have without being rejected as a non-specific candidate target. CasValue reports all input candidates that have not been rejected. All scoring parameters are under user control but default to 1 for each mismatch (for both seed and non- seed), 1 for the alternate PAM cost, and 3 for –x. The ‘accept’ value defaults to 1, the match limit appropriate to specific genome-endogenous targets (where the single accepted match would the endogenous target itself), but may be set to 0 to test exogenous sequence targets, or > 1 for targets that may be occur multiple times in homologs or otherwise similar sequences. 4. CasFinder is passed an input sequence file that may contain either full FASTA sequences, or FASTA headers alone (i.e., without appended sequence) that specify sequence location ranges and the keyword *EXTRACT*: For these, CasFinder extracts the location ranges directly from the genome sequence files